Which of the following model risk management controls is MOST effective at detecting gradual label drift in a supervised classification model?
Select an answer to reveal the explanation.
Short Explanation
You detect gradual label drift by monitoring how model prediction scores line up with actual outcomes over time through backtesting. Retraining can help fix drift later, but it won't tell you when the relationship between features and target has shifted.
Full Explanation
The monitoring approach that compares model prediction scores against actual outcomes over time using backtesting is most effective because label drift means the relationship between features and target variable changes over time, and ongoing comparison to observed outcomes reveals declining calibration or discrimination. Periodic retraining on the full historical dataset may help remedy drift but does not detect it. Feature importance tracking using SHAP values at each inference tracks feature influence, not outcome alignment. Encrypting model predictions before storing them in the data warehouse is a security control.